You are here:
The Use of Random Forest Modeling to Determine Important Watershed and Estuarine Drivers/Pressures/Modulating Factors on Estuarine Eutrophication Response in Northeast Estuaries
Citation:
Latimer, J., M. Pelletier, B. Rashleigh, AND M. Charpentier. The Use of Random Forest Modeling to Determine Important Watershed and Estuarine Drivers/Pressures/Modulating Factors on Estuarine Eutrophication Response in Northeast Estuaries. Eighth Interagency Conference on Research in the Watersheds (ICRW8), Corvallis, OR, June 05 - 08, 2023.
Impact/Purpose:
Human activities in watersheds can contribute excess nutrients to estuaries leading to cultural eutrophication (i.e., surplus carbon levels). Such actions include urban and suburban development, agriculture, and other driving forces. The purpose of this research was to determine the factors that influence estuarine eutrophication condition. Random Forest models (a statistical classification tool) were developed using water quality data from 27 different monitoring sources for thirty-seven watershed-estuary “systems.” Coupled with data on human activities for the associated watersheds, predictions on the relationships between estuary conditions and human drivers and pressures were constructed for northeast US estuaries. These models showed that, for example, drivers such as watershed development and population, and pressures such as nitrogen loading were important in predicting total nitrogen, total chlorophyll, and dissolved oxygen concentrations in the study estuaries.
Description:
Estuaries provide valuable ecosystem services to society, but these services can be compromised by anthropogenic watershed activities. In this work, we identify the important watershed/human drivers and pressures and estuarine modulating factors that increase or diminish eutrophication response in northeast US estuaries. We applied an approach to assimilate data from twenty-eight monitoring sources across the entire northeast US (from Canada to Long Island Sound). Random Forest (RF) modeling identified important anthropogenic watershed drivers/pressures and estuarine modulating factors as predictors of surface total nitrogen (TN), chlorophyll-a (Chl), and bottom dissolved oxygen (DO). Nitrogen and phosphorus loading to the estuaries, % forest in the watershed, % impervious surface, and human population were identified as the most important drivers/pressures of TN, Chl, and DO for the entire region. Water clarity, salinity, water temperature, estuary depth, and tidal range were found to be the most important modulating factors. When examining the southern portion of the study area separately, (from Cape Cod to Long Island Sound, i.e., the northern portion of EPA’s Virginian biogeographic province), watershed anthropogenic factors, such as the nitrogen loading to the estuary, nitrogen loading from developed lands, phosphorus loading from manure operations, % agriculture in the watersheds, stream density, and watershed population were all found to be important for TN and Chl concentrations in the water column. Water clarity and salinity were also identified as important modulating factors in explaining the variance in TN and Chl concentrations in estuaries from this sub-region. This work shows that RF modeling can be used with a combination of water quality data collected from multiple independent monitoring sources along with watershed and estuary oceanographic information to identify the most important watershed drivers/pressures and modulating factors affecting eutrophication response. The results from this analysis can inform watershed and estuary management at regional scales.
